ethicalabs/xLSTM-7b-Polymath
LoRA-style vs ethicalabs/xLSTM-7b-Instruct. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of proj_down, proj_up, proj_up_gate changed, in 32 of 32 layers; norms, biases and embeddings are untouched.
Tensors changed: MLP 96/96, other 0/288, Norms 0/97, Embeddings 0/1, lm_head 0/1